With a plain standard configuration I can confirm behavior of wrk on Ubuntu 16.04 (Linux 4.4.0):
Running without reload during run: $ wrk -d 30 -c 100 -t 20 http://localhsot/ Running 30s test @ http://localhost/ 20 threads and 100 connections Thread Stats Avg Stdev Max +/- Stdev Latency 2.45ms 0.89ms 29.75ms 86.77% Req/Sec 2.07k 279.09 5.32k 78.41% 1241732 requests in 30.10s, 1.45GB read Requests/sec: 41248.07 Transfer/sec: 49.25MB Calling reload one time during run: $ wrk -d 30 -c 100 -t 20 http://localhost/ Running 30s test @ http://localhost/ 20 threads and 100 connections Thread Stats Avg Stdev Max +/- Stdev Latency 5.34ms 8.76ms 91.81ms 85.10% Req/Sec 1.22k 0.86k 5.43k 58.83% 688865 requests in 30.10s, 822.49MB read Socket errors: connect 0, read 39, write 0, timeout 0 Requests/sec: 22887.01 Transfer/sec: 27.33MB Could not verify a connection termination with a reload during a long running request. Might be a "problem" with wrk? Posted at Nginx Forum: https://forum.nginx.org/read.php?2,275328,275329#msg-275329 _______________________________________________ nginx mailing list [email protected] http://mailman.nginx.org/mailman/listinfo/nginx
